Prediction of Spin Polarized Fermi Arcs in Quasiparticle Interference of CeBi

Zhao Huang,1 Christopher Lane,1, 2 Chao Cao,3 Guo-Xiang Zhi,4 Yu Liu,5 Christian E. Matt,5 Brinda Kuthanazhi,6, 7 Paul C. Canfield,6, 7 Dmitry Yarotski,2 A. J. Taylor,2 and Jian-Xin Zhu1, 2, * 1Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA 2Center for Integrated Nanotechnology, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA 3Department of Physics, Hangzhou Normal University, Hangzhou 310036, China 4Department of Physics, Zhejiang University, Hangzhou 310013, China 5Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA 6Ames Laboratory, Iowa State University, Ames, Iowa 50011, USA 7Department of Physics and Astronomy, Iowa State University, Ames, Iowa 50011, USA (Dated: January 15, 2021) We predict that CeBi in the ferromagnetic state is a Weyl semimetal. Our calculations within density func- tional theory show the existence of two pairs of Weyl nodes on the momentum path (0,0,kz) at 15 meV above and 100 meV below the Fermi level. Two corresponding Fermi arcs are obtained on surfaces of mirror- symmetric (010)-oriented slabs at E = 15 meV and both arcs are interrupted into three segments due to hy- bridization with a set of trivial surface bands. By studying the spin texture of surface states, we find the two Fermi arcs are strongly spin-polarized but in opposite directions, which can be detected by spin-polarized ARPES measurements. Our theoretical study of quasiparticle interference (QPI) for a nonmagnetic impurity at the Bi site also reveals several features related to the Fermi arcs. Specifically, we predict that the spin polariza- tion of the Fermi arcs leads to a bifurcation-shaped feature only in the spin-dependent QPI spectrum, serving as a fingerprint of the Weyl nodes.

I. INTRODUCTION cols. Only very recently have Co3Sn2S2 and Co2MnGa been reported as possible magnetic Weyl semimetals with six Weyl 4–6 Despite being predicted more than 90 years ago as an el- nodes and nodal lines, respectively . This is far from enough egant special solution of the relativistic wave equation, Weyl to explore the wide range of possible exotic states. Inter- fermions have never been observed in nature. However, just estingly, recent experimental reports of a pronounced nega- as the sun was setting on the Weyl fermion, they were ob- tive magnetoresistance and observed band inversion in cerium monopnictides35–37 and the geometrically frustrated Shastry- served as a quasiparticle in the electronic structure of TaAs, 38 thereby pushing Weyl physics back to the vanguard and invig- Sutherland lattice GdB4 could provide clues to a possible orating the search for other exotic relativistic particles within new material family harboring Weyl physics. solids. Weyl fermions appear in condensed matter systems when a single Dirac cone is split into two by the breaking of In this paper, we argue that CeBi in the ferromagnetic either time-reversal1–6 or inversion symmetry7–13. This pro- (FM) state is a magnetic Weyl semimetal using first-principles cess generates a pair of cone-like features with opposite chi- LDA+U calculations and we predict robust quasiparticle inter- rality defined by their Berry curvature14. As a direct con- ference (QPI) signatures of spin polarized Fermi arcs directly sequence, anomalous surface states appear connecting Weyl accessible by spin resolved scanning tunneling spectroscopy nodes of opposite topological charge, providing an experi- (STS). By aligning the Ce magnetic moments along the c- mentally accessible fingerprint of the underlying non-trivial axis, we find two pairs of Weyl nodes on the zone diagonal 4,15–21 band topology . along the kz-axis at 15 meV above and 100 meV below the Recently, magnetic Weyl semi-metals have been gaining Fermi level. Then by examining a (010)-oriented slab which interest for providing a new platform to study the inter- is mirror symmetric, two Fermi arcs of opposite spin polar- play between chirality, magnetism, correlation, and topolog- ization are observed on both the top and bottom of surfaces, ical order, thus opening up new routes to novel quantum stretching across the Brillouin zone. These Fermi arcs then states, spin polarized chiral transport22,23, and exotic opti- hybridize with a set of trivial surface states producing a spin cal phenomena24–27. Compared with inversion breaking Weyl vortex encircling the M¯ point in the Brillouin zone. Finally,

arXiv:2008.10628v3 [cond-mat.str-el] 14 Jan 2021 materials, magnetic Weyl materials have the advantage of re- by considering a weak potential scatterer at the Bi site on the ducing the minimum number of allowed Weyl nodes from surface, we predict several features directly related to the scat- four to two. This allows simpler band structures thereby tering among the Fermi arcs in the QPI spectra. Specifically, facilitating clearer comparisons with theoretical predictions, there is a bifurcation-shaped feature originating from the scat- and more robust applications in spintronics28–30 and quantum tering between the two symmetric outer segments of the spin- computation31–34. Another advantage is the magnetic field up Fermi arc. We propose this feature as a direct indicator of tunability which provides us a strong tool on controlling the the Weyl nodes for future experiments. Moreover, due to the band structure and related electromagnetic functionality. isolation of the Weyl nodes near the Fermi level, we find CeBi To date very few magnetic Weyl materials have been syn- to be a robust platform for theoretical and experimental anal- thesized, demonstrating an urgent and important need for ysis of Weyl physics with minimal interference from trivial more theoretical material predictions and robust growth proto- bulk states. 2

II. METHOD

First-principles band structure calculations were carried out within density functional theory framework using the generalized-gradient approximation (GGA) as implemented in the all-electron code WIEN2K39, which is based on the augmented-plane-wave + local-orbitals(APW+lo) basis set. Spin-orbit coupling was included in the self-consistency cy- cles. Furthermore, the Hubbard Coulomb interaction on Ce- 4 f electrons was incorporated to ensure a fully localized Ce-4 f state, which also gives a total magnetic moment of 2µB/Ce consistent with earlier studies oncerium monopnic- tides35,40,41. Although these compounds exhibit several field- FIG. 1. (a) Crystal structure of CeBi with the spin polarization along induced magnetic phases, a ferromagnetic-like state can be the c-axis direction. Here we display a body-centered tetragonal cell stabilized using state-of-the-art cryogenic STM in 3D vector for an otherwise FCC crystal structure of CeBi in bulk. (b) DFT magnetic fields above 4 Tesla35,41,42. The topological analysis electronic band structure of CeBi in the ferromagnetic state with a was performed by employing a real-space tight-binding model red circle highlighting a pair of Weyl nodes near the Fermi level. (c) Same as (b) except a zoom-in around Fermi level along Γ Z with the Hamiltonian, which was obtained by using the wien2wannier − 43 character of various bands indicated. (d) FCC Brillouin zone of FM interface . Bi-6p and Ce-5d states were included in gen- CeBi with the various high-symmetry points marked. The locations erating Wannier functions. It is worth mentioning that, the of the Weyl nodes is given by teal (orange) dots denoting the Weyl DFT+U method for a non-magnetic phase will still leave node as a source (sink) of Berry curvature. f -electron pinned around the Fermi level, which is com- pletely different from the spin-polarized case44. Therefore, the DFT+U approach is not appropriate to describe strongly correlated materials in a paramagnetic state, for which one function composed of 6 Bi-6p and 10 Ce-5d orbitals, and needs to resort to more powerful methods like DFT+DMFT.45 T is the scattering T-matrix containing all multiple scat- To identify the presence of a pair of Weyl nodes in a given tering effects off a single-site impurity. Since the infi- band structure we must check if the band crossings are topo- nite sum of multiple scattering terms form a geometric se- logically trivial or non-trivial. That is, we must calculate ries, the T-matrix can be compactly written as T(ω) = 2 2 1 the Berry curvature at and surrounding the nodal points to I Vimp(2π)− d kG(k,ω) − Vimp, where Vimp is the im- see if the field is divergent or not. To do so, we calculate purity− potential. Here we considered a Born approximation of  R  the Berry curvature Ωn,αβ (k) as implemented in the Wannier- the impurity scattering such that T = Vimp and further chose 46–48 Tools package , for a given band n and momenta k as Vimp to be a constant diagonal matrix with non-zero elements only for p (d) orbitals for a Bi (Ce) impurity site. Our gen- vnm,α (k)vmn,β (k) eral methodology for QPI prediction in CeBi is justified by Ωn,αβ (k) = 2Im ∑ 2 (1) − m=n [εmk εnk] the good agreement between calculated and experimental STS 6 − data on the (001)-oriented slab, shown in detail in the Supple- 50 where εnk is the eigenvalue of Hamiltonian H for band n and mental Material (SM) . momenta k and the matrix elements of the Cartesian velocity operators are given by III. BAND STRUCTURE AND WEYL NODES ∂Hˆ (k) v , (k) = φ φ . (2) nm α h nk ∂k mki α Figure1 (left panel) shows the crystal structure of CeBi in the FM phase along with its corresponding face-centered- Finally the Berry curvature vector is given by cubic (FCC) Brillouin zone, where various high symmetry lines are marked. The middle panel presents the electronic Ωn,γ (k) = εαβγ Ωn,αβ (k) (3) band dispersions evaluated over the full set of high symmetry where εαβγ is the Levi-Civita tensor. points. The energy bands near the Fermi level are mainly of The surface electronic structure was calculated using both Bi-6p and Ce-5d character. Moreover, the moment-carrying the direct diagonalization and the iterative Green’s function localized Ce-4 f states, which provides an effective Zeeman method49,50. The QPI spectra was obtained within the T- exchange field, are seen as a flat band at 3 eV binding ener- matrix approximation as given by: gies. We choose U = 7.9 eV and J = 0.69 eV in our ab initio calculations to make the Ce-4 f energy level consistent with 1 2 2 QPI(q,ω)= i(2π)− d k(2π)− [B(q,ω) B∗( q,ω)],(4) the corresponding state in CeSb as reported by photoemission − − Z spectroscopy51. Moreover, the non-trivial topological nature where B(q,ω) = Tr[G(k + q,ω)T(ω)G(k,ω)] with q and of low-energy states in CeBi are not sensitive to the value of U ω the scattered wave vector and quasiparticle energy, re- once Ce-4 f electrons are quite localized in the ferromagnetic spectively. G is the spin-dependent matrix surface Green’s phase. 3

FIG. 2. Electronic band structure of CeBi in the FM phase along FIG. 3. Surface projected spectral function of the bottom layer of Z Γ Z in the Brillouin zone. The band that passes through the 10-layer (010)-oriented slab (a) and the corresponding bulk (b). −Weyl− nodes− near the Fermi level is shown in red. (b) Berry curvature vector field in the xz-plane for the red band in (a). The Weyl nodes are marked in teal (orange) indicating they are a source (sink) of Berry flux. is more subtle in magnetic Weyl materials. Since the inter- nal magnetic moments on the Ce atoms point in the c-axis direction, the irreducible group elements are different for a (010)-oriented surface compared to those for a (001)-oriented The most important feature of the band structure is the pair surface. Specifically, the (001)-oriented surface breaks the xy of Weyl nodes that appears along Γ Z at the Fermi level. − mirror symmetry and kz is no longer a good quantum num- These two Weyl nodes are formed by the crossing of Bi-6p ber, whereas a (010)- or (100)-oriented surface preserves the and Ce-5d t2g orbital character bands that have been spin split symmetry. Furthermore, a (001)-oriented surface will not re- by the internal Zeeman exchange field [Fig.1 (right panel)]. 4π veal Fermi arcs since the pairs of Weyl nodes with opposite The Weyl nodes w1 and w2 are located at (0,0,0.22) a and chirality are projected to the same point on the surface, thus 4π (0,0,0.28) a in the Brillouin zone and at 15 meV above and cancelling each other, while a (010) or (100)-oriented surface 100 meV below the Fermi level, respectively. Here a is the is expected to produce the Fermi arcs. Here we focus on the lattice constant of the cubic conventional cell and hereafter (010)-oriented surface and examine the resulting electronic we will choose 4π/a = 1. Due to mirror symmetry in the xy- structure and effect of the Fermi arcs on the QPI spectrum. plane, a duplicate pair of Weyl nodes are found along the k − z To distinguish the topologically non-trivial surface states axis, named asw ¯1 andw ¯2. The locations of these four Weyl and the trivial surface bands derived from the bulk, we con- nodes are marked in the Brillouin zone [Fig.1 (left panel)] by sider the band structure of a 10-layer (010)-oriented slab, and the teal (orange) dots, where the teal (orange) color indicates a also the bulk with supercell containing 10 layers of CeBi along Weyl node as a source (sink) of Berry curvature. Importantly, the same direction. The spectral functions for the bottom layer since the Weyl nodes in CeBi are isolated near the Fermi level, of both cases are presented in Fig.3. By comparing the two the experimental signatures and theoretical analysis will be panels, we can see that all bands in Fig.3(a) except for the less obstructed as compared to Co3Sn2S2 and Co2MnGa. Fermi arcs [highlighted in the red frame] have the correspond- Figure2(a) shows w1 and w2 near the Fermi level along the ing bulk bands in Fig.3(b). This indicates that the Fermi arcs Z Γ Z cut in the Brillouin zone. The band that passes − − − are the non-trivial topology induced edge states. The mass- through the Weyl nodes is marked in red. Figure2(b) shows less energy bands corresponding to Fermi arcs are shown in the corresponding Berry curvature vector field in the xz-plane the supplementary50. of the first Brillouin zone for the red band in Fig.2(a). Two Figure4(a) shows the surface spectral function Asurf(k,ω) clear divergences in the vector field are seen along Z Γ ± − for an effective 64-layer (010)-oriented slab in the surface re- at the Weyl nodes, with one node as a source (teal) and the ciprocal space, centered at the M¯ point, for a constant-energy other a sink (orange) of topological charge. This confirms the cut of ω corresponding to w1. Two clear Fermi arcs are seen topologically non-trivial nature of these Weyl band crossings. connecting projections of the Weyl nodes w1 tow ¯1 and w2 to Furthermore, we identified 16 additional pairs of Weyl node w¯2, labeled as arc I and arc II, respectively. By breaking down existing above the Fermi energy, each marked in Fig.2(a) by the spectral function into its orbital components, we find the an orange or teal circle denoting its topological charge. Fermi arcs to be mainly composed of Bi px and py, and Ce t2g orbitals. The dominant Bi px and weaker Bi pz contribution, along with their spin dependence, is given in Fig.4(b). IV. SURFACE SPECTRAL FUNCTIONS Interestingly, as the two Fermi arcs traverse across the Bril- louin zone they are intersected by and hybridize with two Due to the presence of Weyl nodes near the Fermi level we majority pz character trivial bands extending along the kx di- expect the emergence of a pair of Fermi arcs connecting nodes rection [Fig.4(a)]. The avoided crossing between these two of opposite topological charge when CeBi is cleaved along a sets of bands yields a nearly continuous outer cross shaped given crystal plane exposing a surface. However, the situation Fermi surface and a rectangular pocket surrounding M¯ . More- 4

FIG. 4. ((a) Surface spectral function of a (010)-oriented surface of FM CeBi. The dash lines draw the boundary of the surface first Bril- louin zone. I and II denotes the two Fermi arcs connecting the Weyl nodes. (b) Spin- and orbital-resolved surface spectral functions. (c) FIG. 5. (a) QPI spectra for a (010)-oriented surface of FM CeBi Spin polarization Sz in the Brillouin zone. (d) Schematic of the spin texture of varioush i Fermi arcs in the Brillouin zone. (e) Calcu- with the dominant scattering vectors indicated. (b) Surface spectral lated spin texture of the Fermi arcs. function with the trivial pz states removed with various characteristic scattering transitions marked. (c) QPI spectra corresponding to the surface spectral function in (b). (d) Surface spectral function with Fermi arcs and spin vortex removed, with dominant scattering vec- tors indicated. (e) QPI spectra corresponding to the surface spectral over, additional trivial surface states are observed around the function in (d). QPI spectra of the spin-up (f) and spin-down channel Γ¯ point, but they do not interact with the arcs. (g) corresponding to the surface spectral function in Fig.4(a). Figure4(c) shows s (k,ω) = 1 ImTr[σ G(k,ω)] in the h z i − π 3 first Brillouin zone for a constant-energy cut of ω = w1, where red (blue) indicates the strength of the positive (negative) σ3 projection of the surface spectral function. Here we find arc I and arc II to be oppositely spin polarized, which is also reflected in Fig.4(b). This is the consequence of the spin- ¯ lies completely in the (k ,k ) plane, where s is strictly zero. momentum locking effect around the M point which origi- z x h yi nates from the strong spin-orbital coupling in CeBi. More- The absence of any out-of-plane spin polarization Sy is guar- over, the strength of polarization is generically greater for arc anteed by the combined symmetry of time-reversal symmetry I compared to arc II, due to the FM order along the c-axis. T and twofold rotational symmetry along the y axis C2y, even Figure4(d) and (e) show the full spin texture of the Fermi though T or C2y itself is broken. This symmetry is satisfied the 1 1 arcs in the (010)-oriented slab. We note several important fea- Hamiltonian of the (010) slab with C2yTH(kx,kz)T − C2−y = tures: (i) Arc I is strongly polarized along +z, whereas arc H(kx,kz). Since T flips all spin components and reverses all II is strongly polarized along z. (ii) Due to inversion sym- crystal momenta, and C2y only flips Sx,Sz and reverses kz,kx, − metry, the spin polarization for a given (kz,kx) on the bottom the combined transformation C2yT maps (kz,kx) to itself and 2 surface is equivalent to the spin polarization on the top sur- reverses Sy. Because (C2yT) = 1, there is no Kramers de- face at ( k , k ). (iii) The Fermi surface pocket surround- generacy, implying C T k ,k and k ,k can only differ by − z − x 2y | z xi | z xi ing M¯ forms a spin vortex in the momentum space, owing to a phase, forbidding different Sy. Therefore, the expectation strong spin-orbit coupling. (iv) The spin-polarization vector value of Sy must be strictly zero throughout the Brillouin zone. 5

FIG. 6. (a) Zoom-in view of the QPI spectra in Fig.5(a) highlighting the bifurcated feature corresponding to direct Fermi arc-Fermi arc scattering. (b) Close-up of the surface spectral function in Fig.4(a) with scattering vectors Q0 and Q” connecting the two outer segments of the spin-up Fermi arc.

V. QUASIPARTICLE INTEREFERENCE FIG. 7. Spectral function around one Fermi arc at (a) 50 meV, (b) 0 meV and (c) 50 meV. QPI spectra in the same regime− as Fig.4(a) Bi-site impurity – At first we present our theoretical study at (d) 50 meV, (e) 0 meV and (f) 50 meV. on the (010)-oriented surface QPI derived from scattering be- − tween these surface states due to an isotropic nonmagnetic impurity at Bi site. Figure5(a) shows the QPI spectra in the first Brillouin zone for a constant energy cut of ω = w1. fers momentum between the Fermi pocket to the trivial bands Firstly, we notice that almost all q = 0 scattering is sup- around Γ¯ . If we now consider the spectral function in the re- pressed by destructive interference between the various multi- gion away from the Fermi arcs and spin vortex [Fig.5(d)], we orbital scattering channels. This phenomena is not present in find the remaining Fermi sheets clearly produce the remaining the simple joint density of state description since the phase features Q4 and Q5 [Fig.5(e)]. Q4 comes from the internal of wavefunctions is neglected. But because QPI(0,ω) ∝ scattering of the bands around Γ¯ , whereas the Q5 scattering 2 d kImTr[G(k,ω)T(k,ω)G(k,ω)], the off diagonal compo- vector connects pz dominant states to those surrounding Γ¯ . nents of the matrix Green’s function contribute to the QPI sig- R Since the surface spectral function exhibits a strong spin nal. Moreover, the sign of the imaginary part of the product texture, it is reasonable to expect a highly spin dependent QPI. is highly sensitive to the relative sign between real and imag- Figures5(f)-(g) show the up and down-spin projection of the inary parts of the Green function and k in the Brillouin zone. QPI spectra. These two patterns are quite similar except for Therefore, even after integrating over the full zone the results scattering momenta related to Q1. Since Q1 connects Fermi may be quite small. A similar cancellation takes place for arcs of the same +z spin polarization, the intensity in the spin- scattering at finite q, which leads to a quite broad and smooth up QPI channel is stronger than that in the spin-down channel. QPI spectrum despite sharp features in the spectral function. This stark difference in the spin-up and spin-down QPI maps In Fig.5(a) the five main features in the QPI pattern, cor- provides a direct experimentally accessible prediction of the responding to strong scattering between the surface states, are underlying spin-polarized Fermi arcs. labeled with Q1 through Q5 and are highlighted by red dash Thus far, we have covered all the strong scattering transi- circles and rectangles. In order to pinpoint the origin of the tions in the QPI map. However, the scattering between the these strong features, we isolate various sections of the surface two outer Fermi arcs appears to be missing. There should be spectral function by masking out the Green function in other features in the QPI spectra corresponding to this scattering sections and amplify their effect on the QPI spectra. If we channel. Indeed, upon close inspection, a red flat tail can be only consider the scattering between the Fermi arcs, spin vor- seen at (q ,q ) = ( 0.25,0) in Figs.5(a),5(c) and5(f) cor- z x ± tex, and bands surrounding Γ¯ [Fig.5(b)] we find the Q1, Q2 responding to this ‘missing’ scattering pathway. Figure6(a) and Q3 features persist in the QPI [Fig.5(c)]. Therefore, we shows a zoomed in section of the QPI map given in Fig.5(a) find the momentum transfer Q1 to connect the outer segment displaying a dispersing bifurcating feature. The trunk, which of the Fermi arc to the pocket surrounding M¯ . Q2 scatters is the tail mentioned above, extends from qz = 0.23 to about electrons from the outer segments of Fermi arcs to the trivial qz = 0.26, and bifurcates into two branches. The intensity bands around Γ¯ . Because the Fermi arc is curved, a curved of the two branches decreases with increasing q , but the | z| bridge-like dispersion is seen in the Q2 rectangle. Q3 trans- branch between qz = 0.26 to qz = 0.33 is still clearly seen. 6

FIG. 8. Spectral functions of (a) spin-up dxy, (b) spin-down dxy, (c) spin-up dxz, and (d) spin-down dxz. (e) total, (f) spin-up and (g) spin-down QPI spectrum.

This bifurcation shaped feature originates from electrons scat- those found for a Bi-site scatter because the existence of a tered between the shoulder-like portion (denoted by Q0) and dense set of Ce-5d bands surrounding the Γ¯ point at the Fermi the slope (marked by Q00) of the Fermi arc, as shown in Fig. level that generate a significant trivial band scattering back- 6(b). The enhanced scattering along qx = 0 originates from ground, thus, washing out most features related to Fermi arc pure shoulder-shoulder Q0 scattering spanning between the scattering in the QPI. mirrored sections of the Fermi arcs. For qz > 0.26, Q00 mo- mentum transfers becomes dominant, tracing out the curva- ture of the Fermi arc. The intensity of the QPI decreases in- creased Q00 due to the reduction in the nesting between the shoulder and the curving Fermi arc, therefore, generating the VI. QPI FOR HIGHER ENERGY WEYL NODE two bifurcating branches in the QPI spectrum. Figure7 shows the energy dependence of the Fermi arc I In the previous sections, we have studied the surface states and the bifurcation feature. From panel (a)-(c), we see that and QPI spectra originating from near the Fermi level of a as energy increases from 50meV to 50meV, the Fermi arc (010)-oriented slab. Here, we study the surface projected becomes more curved due− to the energetic distance from the spectral function and QPI map of the (010)-oriented slab for Weyl node50. This manifests in QPI as a growing angle of the a pair of Weyl nodes above the Fermi level at 0.915 eV, as bifurcation branches as shown in Fig.7(d)-(f). indicated by the red dashed line in Fig.9(a). We chose this Ce-site impurity – Now we replace the Bi impurity with a specific pair of Weyl nodes since they are quite isolated from Ce impurity and mark the difference in the surface spectral other intruding bands, which tend to complicate the observa- function and QPI. Here, the d-orbitals play the key role in tion of the Fermi arcs. This Weyl node and the one just below the Ce impurity induced scattering. We find two noteworthy it are both sources of Berry flux as indicated in Fig.2(a). The scattering vectors labeled Q1 and Q2 shown in Fig.8. Q1 con- surface spectral function at this energy is shown by Fig.9(b), nects the Fermi arcs with trivial bands around Γ. Addition- where two clear wing-like features are found near Γ along the ally, because the Ce magnetic moments are along the posi- kx axis. Zooming-in on these features [Fig.9(c)], it is easy tive z direction, the spectral function for spin-up d orbitals are to see that they are Fermi arcs which connect the Weyl nodes. more intense than the spin-down d orbitals. As a result, the Figure9(d) shows the σ3 projection of the spectral function Q1 spike-like feature in the total and spin-up QPI spectrum is Sz in the first Brillouin zone, where the red (blue) intensity shown in Fig.8(e) and8(f), whereas this feature is not clear indicatesh i the strength of the spin-up (spin-down) polarization. in Fig.8(g). The momentum transfer Q2, which is similar to Figure9(e) shows a close-up showing a clear spin texture of Q4 in the case of Bi impurity, comes from the internal scatter- the Fermi arc. Similar to our earlier discussion, the scattering ing of the bands around the Γ point. This scattering channel is between the Fermi arcs with different spin textures can lead strong for both spin-up and spin-down channels, which makes to interesting features in the total and spin resolved QPI spec- it a strong feature in all QPI maps. tra, as shown in Fig.9(f),9(g) and9(h). Because the spin-up The features displayed for a Ce impurity are not as rich as Fermi arcs are curved and the spin-down arcs are relatively 7

FIG. 9. (a) Theoretical electronic band structure around the Weyl nodes at 0.915 eV. Surface projected spectral function for ω = 0.915 eV in (b) the first Brillouin zone and (c) zoom-in around Γ showing the Fermi arcs. Sz in (d) the first Brillouin zone and (e) zoom-in around Γ. (f) total, (g) spin-up and (h) spin-down QPI spectra. h i

flat, the QPI map yields curved and flat features in the spin- and theoretical studies may be initiated to examine the other up and spin-down QPI spectra, respectively, as indicated by aspects of Weyl physics in CeBi and other magnetic Weyl ma- the black arrows. The total QPI also displays a similar curved terials. feature as shown in Fig.9(f). ACKNOWLEDGMENTS

VII. CONCLUSION Acknowledgments – We thank Jenny Hoffman for use- ful discussions. This work was carried out under the In summary, we have found theoretically that CeBi is a auspices of the U.S. Department of Energy (DOE) Na- magnetic Weyl semimetal with two pairs of Weyl nodes close tional Nuclear Security Administration under Contract No. to the Fermi level. The induced two Fermi arcs have op- 89233218CNA000001. It was supported by the Center for the posite spin polarization directions on each surface of a mir- Advancement of Topological Semimetals, a DOE BES EFRC. ror symmetric slab. A nonmagnetic impurity at Bi site can C.L. was supported by LANL LDRD Program. Additional lead to quasiparticle interference pattern with features arising support was provided in part by the Center for Integrated Nan- from the Fermi arcs. There is a novel bifurcation shaped fea- otechnologies, a DOE BES user facility, in partnership with ture available in spin-up but absent in spin-down QPI pattern, the LANL Institutional Computing Program for computa- which is a unique property of this material and can be veri- tional resources. C.E.M. was supported by the Swiss National fied with spin-resolved scanning tunneling probes. Our cal- Science Foundation under fellowship No. P2EZP2 175155 culations provide a baseline, from which further experimental and P400P2 183890.

* [email protected] face states in the electronic structure of pyrochlore iridates,” Phys. 1 Xiangang Wan, Ari M. Turner, Ashvin Vishwanath, and Rev. B 83, 205101 (2011). Sergey Y. Savrasov, “Topological semimetal and Fermi-arc sur- 8

2 Gang Xu, Hongming Weng, Zhijun Wang, Xi Dai, and Zhong of a Weyl semimetal,” Science 351, 1184–1187 (2016), Fang, “Chern semimetal and the quantized anomalous hall effect https://science.sciencemag.org/content/351/6278/1184.full.pdf. 17 in HgCr2Se4,” Phys. Rev. Lett. 107, 186806 (2011). Hao Zheng, Su-Yang Xu, Guang Bian, Cheng Guo, Guoqing 3 Zhijun Wang, M. G. Vergniory, S. Kushwaha, Max Hirschberger, Chang, Daniel S. Sanchez, Ilya Belopolski, Chi-Cheng Lee, Shin- E. V. Chulkov, A. Ernst, N. P. Ong, Robert J. Cava, and B. An- Ming Huang, Xiao Zhang, Raman Sankar, Nasser Alidoust, Tay- drei Bernevig, “Time-reversal-breaking Weyl fermions in mag- Rong Chang, Fan Wu, Titus Neupert, Fangcheng Chou, Horng- netic heusler alloys,” Phys. Rev. Lett. 117, 236401 (2016). Tay Jeng, Nan Yao, Arun Bansil, Shuang Jia, Hsin Lin, and 4 Noam Morali, Rajib Batabyal, Pranab Kumar Nag, Enke Liu, Qi- M. Zahid Hasan, “Atomic-scale visualization of quantum inter- unan Xu, Yan Sun, Binghai Yan, Claudia Felser, Nurit Avraham, ference on a Weyl semimetal surface by scanning tunneling mi- and Haim Beidenkopf, “Fermi-arc diversity on surface termina- croscopy,” ACS Nano 10, 1378–1385 (2016), pMID: 26743693, tions of the magnetic Weyl semimetal Co3Sn2S2,” Science 365, https://doi.org/10.1021/acsnano.5b06807. 1286–1291 (2019). 18 Rajib Batabyal, Noam Morali, Nurit Avraham, Yan Sun, 5 D. F. Liu, A. J. Liang, E. K. Liu, Q. N. Xu, Y. W. Li, C. Chen, Marcus Schmidt, Claudia Felser, Ady Stern, Binghai Yan, D. Pei, W. J. Shi, S. K. Mo, P. Dudin, T. Kim, C. Cacho, G. Li, and Haim Beidenkopf, “Visualizing weakly bound sur- Y. Sun, L. X. Yang, Z. K. Liu, S. S. P. Parkin, C. Felser, and face Fermi arcs and their correspondence to bulk Weyl Y. L. Chen, “Magnetic Weyl semimetal phase in a kagome´ crys- fermions,” Science Advances 2 (2016), 10.1126/sciadv.1600709, tal,” Science 365, 1282–1285 (2019). https://advances.sciencemag.org/content/2/8/e1600709.full.pdf. 6 Ilya Belopolski, Kaustuv Manna, Daniel S. Sanchez, Guoqing 19 Paolo Sessi, Yan Sun, Thomas Bathon, Florian Glott, Zhilin Li, Chang, Benedikt Ernst, Jiaxin Yin, Songtian S. Zhang, Tyler Hongxiang Chen, Liwei Guo, Xiaolong Chen, Marcus Schmidt, Cochran, Nana Shumiya, Hao Zheng, Bahadur Singh, Guang Claudia Felser, Binghai Yan, and Matthias Bode, “Impurity Bian, Daniel Multer, Maksim Litskevich, Xiaoting Zhou, Shin- screening and stability of Fermi arcs against coulomb and mag- Ming Huang, Baokai Wang, Tay-Rong Chang, Su-Yang Xu, Arun netic scattering in a Weyl monopnictide,” Phys. Rev. B 95, 035114 Bansil, Claudia Felser, Hsin Lin, and M. Zahid Hasan, “Discov- (2017). ery of topological Weyl fermion lines and drumhead surface states 20 Andras´ Gyenis, Hiroyuki Inoue, Sangjun Jeon, Brian B Zhou, in a room temperature magnet,” Science 365, 1278–1281 (2019). Benjamin E Feldman, Zhijun Wang, Jian Li, Shan Jiang, Quinn D 7 Hongming Weng, Chen Fang, Zhong Fang, B. Andrei Bernevig, Gibson, Satya K Kushwaha, Jason W Krizan, Ni Ni, Robert J and Xi Dai, “Weyl semimetal phase in noncentrosymmetric Cava, B Andrei Bernevig, and Ali Yazdani, “Imaging electronic transition-metal monophosphides,” Phys. Rev. X 5, 011029 states on topological semimetals using scanning tunneling mi- (2015). croscopy,” New Journal of Physics 18, 105003 (2016). 8 Su-Yang Xu, Ilya Belopolski, Nasser Alidoust, Madhab Neu- 21 Guoqing Chang, Su-Yang Xu, Hao Zheng, Chi-Cheng Lee, Shin- pane, Guang Bian, Chenglong Zhang, Raman Sankar, Guoqing Ming Huang, Ilya Belopolski, Daniel S. Sanchez, Guang Bian, Chang, Zhujun Yuan, Chi-Cheng Lee, Shin-Ming Huang, Hao Nasser Alidoust, Tay-Rong Chang, Chuang-Han Hsu, Horng-Tay Zheng, Jie Ma, Daniel S. Sanchez, BaoKai Wang, Arun Ban- Jeng, Arun Bansil, Hsin Lin, and M. Zahid Hasan, “Signatures of sil, Fangcheng Chou, Pavel P. Shibayev, Hsin Lin, Shuang Jia, Fermi arcs in the quasiparticle interferences of the Weyl semimet- and M. Zahid Hasan, “Discovery of a Weyl fermion semimetal als TaAs and NbP,” Phys. Rev. Lett. 116, 066601 (2016). and topological Fermi arcs,” Science 349, 613–617 (2015), 22 Pavan Hosur and Xiaoliang Qi, “Recent developments in transport https://science.sciencemag.org/content/349/6248/613.full.pdf. phenomena in Weyl semimetals,” Comptes Rendus Physique 14, 9 B. Q. Lv, H. M. Weng, B. B. Fu, X. P. Wang, H. Miao, J. Ma, 857 – 870 (2013). P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, Z. Fang, 23 Fernando de Juan, Adolfo G. Grushin, Takahiro Mori- X. Dai, T. Qian, and H. Ding, “Experimental discovery of Weyl moto, and Joel E Moore, “Quantized circular photogal- semimetal TaAs,” Phys. Rev. X 5, 031013 (2015). vanic effect in Weyl semimetals,” Nat. Commun. 8 (2017), 10 Z. K. Liu, L. X. Yang, and et. al., “Evolution of the fermi sur- https://doi.org/10.1038/ncomms15995 (2017). face of Weyl semimetals in the transition metal pnictide family,” 24 Takahiro Morimoto, Shudan Zhong, Joseph Orenstein, and Nature Mater. 15, 27 (2016). Joel E. Moore, “Semiclassical theory of nonlinear magneto- 11 J. Jiang, Z. Liu, Y. Sun, and et al, “Signature of type-II weyl optical responses with applications to topological Dirac/Weyl semimetal phase in MoTe2,” Nat. Commun. 8, 13973 (2017). semimetals,” Phys. Rev. B 94, 245121 (2016). 12 L. Huang, T. M. McCormick, and et al, “Spectroscopic evidence 25 A. A. Zyuzin and A. Yu. Zyuzin, “Chiral anomaly and second- for a type II weyl semimetallic state in MoTe2,” Nat. Mater. 15, harmonic generation in Weyl semimetals,” Phys. Rev. B 95, 1155 (2016). 085127 (2017). 13 A. Tamai, Q. S. Wu, I. Cucchi, F. Y. Bruno, S. Ricco,` T. K. 26 N. Sirica, R. I. Tobey, L. X. Zhao, G. F. Chen, B. Xu, R. Yang, Kim, M. Hoesch, C. Barreteau, E. Giannini, C. Besnard, A. A. B. Shen, D. A. Yarotski, P. Bowlan, S. A. Trugman, J.-X. Zhu, Soluyanov, and F. Baumberger, “Fermi arcs and their topological Y. M. Dai, A. K. Azad, N. Ni, X. G. Qiu, A. J. Taylor, and character in the candidate type-II Weyl semimetal MoTe2,” Phys. R. P. Prasankumar, “Tracking ultrafast photocurrents in the Weyl Rev. X 6, 031021 (2016). semimetal TaAs using THz emission spectroscopy,” Phys. Rev. 14 N. P. Armitage, E. J. Mele, and Ashvin Vishwanath, “Weyl and Lett. 122, 197401 (2019). Dirac semimetals in three-dimensional solids,” Rev. Mod. Phys. 27 N. Sirica, P. P. Orth, and et. al., “Photocurrent-driven 90, 015001 (2018). transient symmetry breaking in the Weyl semimetal TaAs,” 15 A. V. Balatsky, I. Vekhter, and Jian-Xin Zhu, “Impurity-induced arXiv:2005,10308. states in conventional and unconventional superconductors,” Rev. 28 A. A. Zyuzin and A. A. Burkov, “Topological response in Weyl Mod. Phys. 78, 373–433 (2006). semimetals and the chiral anomaly,” Phys. Rev. B 86, 115133 16 Hiroyuki Inoue, Andras´ Gyenis, Zhijun Wang, Jian (2012). Li, Seong Woo Oh, Shan Jiang, Ni Ni, B. Andrei 29 M. M. Vazifeh and M. Franz, “Electromagnetic response of Weyl Bernevig, and Ali Yazdani, “Quasiparticle interfer- semimetals,” Phys. Rev. Lett. 111, 027201 (2013). ence of the Fermi arcs and surface-bulk connectivity 9

30 Qiang Li, Dmitri E. Kharzeev, and et. al., “Chiral magnetic effect 41 Brinda Kuthanazhi, Na Hyun Jo, Li Xiang, Sergey L Bud’ko, in ZrTe5,” Nature Phys. 12, 550 (2016). and Paul C. Canfield, “Metamagnetism and magnetoresistance in 31 Tobias Meng and Leon Balents, “Weyl superconductors,” Phys. CeBi single crystals,” arXiv.1912.08896 (2019). Rev. B 86, 054504 (2012). 42 C. Trainer, C. M. Yim, M. McLaren, and P. Wahl, “Cryo- 32 Pavan Hosur, Xi Dai, Zhong Fang, and Xiao-Liang Qi, “Time- genic stm in 3d vector magnetic fields realized through a rotat- reversal-invariant topological superconductivity in doped Weyl able insert,” Review of Scientific Instruments 88, 093705 (2017), semimetals,” Phys. Rev. B 90, 045130 (2014). https://doi.org/10.1063/1.4995688. 33 He Wang, Huichao Wang, Yuqin Chen, Jiawei Luo, Zhujun Yuan, 43 Jan Kunes,ˇ Ryotaro Arita, Philipp Wissgott, Alessandro Toschi, Jun Liu, Yong Wang, Shuang Jia, Xiong-Jun Liu, Jian Wei, and Hiroaki Ikeda, and Karsten Held, “Wien2wannier: From lin- Jian Wang, “Discovery of tip induced unconventional supercon- earized augmented plane waves to maximally localized wannier ductivity on Weyl semimetal,” Science Bulletin 62, 425 – 430 functions,” Computer Physics Communications 181, 1888 – 1895 (2017). (2010). 34 Zhongbo Yan, Zhigang Wu, and Wen Huang, “Vortex end majo- 44 Roxanne M. Tutchton, Wei-ting Chiu, R. C. Albers, G. Kotliar, rana zero modes in superconducting Dirac and Weyl semimetals,” and Jian-Xin Zhu, “Electronic correlation induced expansion of Phys. Rev. Lett. 124, 257001 (2020). fermi pockets in δ-plutonium,” Phys. Rev. B 101, 245156 (2020). 35 C. Guo, C. Cao, and M. et al. Smidman, “Possible Weyl fermions 45 Dong-Choon Ryu, Junwon Kim, Kyoo Kim, Chang-Jong Kang, in the magnetic kondo system CeSb,” npj Quant Mater 2, 39 J. D. Denlinger, and B. I. Min, “Distinct topological properties in (2017). ce monopnictides having correlated f electrons: CeN vs. CeBi,” 36 Kenta Kuroda, M. Ochi, H. S. Suzuki, M. Hirayama, Phys. Rev. Research 2, 012069 (2020). M. Nakayama, R. Noguchi, C. Bareille, S. Akebi, S. Kunisada, 46 QuanSheng Wu, ShengNan Zhang, Hai-Feng Song, Matthias T. Muro, M. D. Watson, H. Kitazawa, Y. Haga, T. K. Kim, Troyer, and Alexey A. Soluyanov, “WannierTools : An open- M. Hoesch, S. Shin, R. Arita, and Takeshi Kondo, “Experimental source software package for novel topological materials,” Com- determination of the topological phase diagram in cerium monop- put. Phys. Commun. 224, 405–416 (2018), arXiv:1703.07789 nictides,” Phys. Rev. Lett. 120, 086402 (2018). [physics.comp-ph]. 37 Paul C. Canfield, “New materials physics,” Rep. Prog. Phys. 83, 47 Xinjie Wang, Jonathan R. Yates, Ivo Souza, and David Vander- 016501 (2020). bilt, “Ab initio calculation of the anomalous hall conductivity by 38 W. Shon, D.-C. Ryu, K. Kim, B.I. Min, B. Kim, B. Kang, B.K. wannier interpolation,” Phys. Rev. B 74, 195118 (2006). Cho, H.-J. Kim, and J.-S. Rhyee, “Magnetic field–induced type 48 Di Xiao, Ming-Che Chang, and Qian Niu, “Berry phase effects II Weyl semimetallic state in geometrically frustrated Shastry- on electronic properties,” Rev. Mod. Phys. 82, 1959–2007 (2010). 49 Sutherland lattice GdB4,” Materials Today Physics 11, 100168 M P Lopez Sancho, J M Lopez Sancho, J M L Sancho, and J Ru- (2019). bio, “Highly convergent schemes for the calculation of bulk and 39 Peter Blaha, Karlheinz Schwarz, Georg K H Madsen, Dieter surface green functions,” Journal of Physics F: Metal Physics 15, Kvasnicka, Joachim Luitz, Robert Laskowsk, Fabien Tran, Lau- 851–858 (1985). rence Marks, and Laurence Marks, “Wien2k: An augmented 50 “See supplemental material for more details,” . plane wave + local orbitals program for calculating crystal proper- 51 Sooyoung Jang, Robert Kealhofer, Caolan John, Spencer ties (Karlheinz Schwarz Techn. Universitat¨ Wien, 2019),” (2019). Doyle, Ji-Sook Hong, Ji Hoon Shim, Q. Si, O. Erten, 40 V. P. Liechtenstein, A. I. and. Antropov and B. N. Harmon, “Elec- Jonathan D. Denlinger, and James. G. Analytis, “Di- tronic structure and magneto-optical effects in CeSb,” Phys. Rev. rect visualization of coexisting channels of interaction in B 49, 10770 (1994). CeSb,” Science Advances 5 (2019), 10.1126/sciadv.aat7158, https://advances.sciencemag.org/content/5/3/eaat7158.full.pdf. Supplemental Material for "Prediction of Spin Polarized Fermi Arcs in Quasiparticle Interference of CeBi"

Zhao Huang,1 Christopher Lane,1, 2 Chao Cao,3 Guo-Xiang Zhi,4 Yu Liu,5 Christian E. Matt,5 6, 7 6, 7 2 2 1, 2, Brinda Kuthanazhi, Paul C. Canfield, Dmitry Yarotski, A. J. Taylor, and Jian-Xin Zhu ∗ 1Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA 2Center for Integrated Nanotechnology, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA 3Department of Physics, Hangzhou Normal University, Hangzhou 310036, China 4Department of Physics, Zhejiang University, Hangzhou 310013, China 5Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA 6Ames Laboratory, Iowa State University, Ames, Iowa 50011, USA 7Department of Physics and Astronomy, Iowa State University, Ames, Iowa 50011, USA (Dated: January 15, 2021)

FORMULA OF QUASIPARTICLE INTERFERENCE

In this section we derive the expression for the quasiparticle interference for the surface layer of a slab with two atoms A and B in a primitive cell. The location of all atoms on the surface is given by an in-plane vector ri, where 푖 even denotes the A sublattice and 푖 odd gives the B sublattice. The Green’s function of the entire slab with an impurity at site r1 (type B) is given by

Σ 푗 푖휔푛훿푖 푗 I h G j0, j = 훿 I V 훿푖 G 1, j , (S1) 0 [ 0 − ij0 ] ( ) ij + imp 1 ( )

where 푖, 푗, 푗 0 index the atom locations throughout the entire slab, hij0 is the hopping matrix connecting orbitals in unit cell 푖 to those in unit cell 푗 0, and Vimp is the impurity potential matrix. Moreover, Vimp is block diagonal in atomic species, with only a 6 by 6 nonzero block for B-type vacancy only composed of 3 Bi-6푝 orbitals plus spin. This means if the A site is composed of 5 Ce-5푑 orbitals plus spin degrees of freedom, the overall dimension of I, h, G and Vimp can grow rapidly for increased number of layers. Then by defining the Green’s function in the pristine case

Σ 푗 푖휔푛훿푖 푗 I h G j0, j = 훿 I, (S2) 0 [ 0 − ij0 ] 0 ( ) ij

we can write the full greens function in terms of the Timp matrix,

G i, j = G i, j G i, 1 T G 1, j , (S3) ( ) 0 ( ) + 0 ( ) imp 0 ( )

where Timp is obtained as

1 T = V 퐼 G 1, 1 V − . (S4) imp imp ( − 0 ( ) imp)

If we only consider the first Born approximation, where 푉푖푚푝 1, the deviations from G can be written as  0 훿G i, i = G i, 1 V G 1, i . (S5) ( ) 0 ( ) imp 0 ( ) This determines the corresponding local density of states for sites A and B, 1 휌 2푖 = ImTr G 2i, 1 V G 1, 2i , (S6) ( ) − 휋 [ 0 ( ) imp 0 ( )] 1

arXiv:2008.10628v3 [cond-mat.str-el] 14 Jan 2021 휌 2푖 1 = ImTr G 2i 1, 1 V G 1, 2i 1 . (S7) ( + ) − 휋 [ 0 ( + ) imp 0 ( + )]

Since we only wish to examine the QPI spectrum relevant to STM measurements, we restrict G0 to just surface atomic sites. Therefore, we define the surface local density of states for sites A and B as 1 휌푠 2푖 = ImTr G 2i, 1 V G 1, 2i , (S8) ( ) − 휋 [ s0 ( ) p s0 ( )] 1 휌푠 2푖 1 = ImTr G 2i 1, 1 V G 1, 2i 1 , (S9) ( + ) − 휋 [ s0 ( + ) p s0 ( + )] 2 where 퐺푠 is the surface Green function with a dimension of 10 6 or 6 10 6 6 for site A (B), and V is the nonzero 6 6 0 × × ( × ) p × block of Vimp. Since we have periodic boundary conditions in the 푥푦 plane of the slab geometry, we are able to perform Fourier transform to obtain the quasiparticle interference spectral function at the surface as

휌푠퐴 q = 휌푠 2푖 exp 푖q r2i , (S10) ( ) 푖 ( ) (− · ) ∑︁ dp dp G 2i, 1 = 푑 푖 푝† = G k exp 푖k r r with G k = 푑 푝† , (S11) s0 ( ) h 2 1i s0 ( ) ( ·( 2i − 1)) s0 ( ) h k ki ∑︁k pd pd G 1, 2i = 푝 푑† = G k exp 푖k r r with G k = 푝 푑† , (S12) s0 ( ) h 1 2푖i s0 ( ) (− ·( 2i − 1)) s0 ( ) h k ki ∑︁k 휌푠퐵 q = 휌푠 2푖 1 exp 푖q r2i 1 , (S13) + ( ) 푖 ( + ) (− · ) ∑︁ pp pp Gs0 2i 1, 1 = 푝2푖 1 푝† = G k exp 푖k r2i 1 r1 with G k = 푝k 푝† , (S14) ( + ) h + 1i s0 ( ) [ ·( + − )] s0 ( ) h ki ∑︁k pp Gs0 1, 2i 1 = G k exp 푖k r2i 1 r1 . (S15) ( + ) s0 ( ) [− ·( + − )] ∑︁k Combining Eqs. (S8), (S10), (S11) and (S12), we arrive at 1 휌푠퐴 q = ImTr G 2i, 1 V G 1, 2i exp 푖q r (S16) 휋 s0 p s0 2i ( ) 푖 − [ ( ) ( )] (− · ) ∑︁ 푖 dp pd = Tr G k V G k0 exp 푖 k k0 r r 푐.푐. exp 푖q r (S17) 휋 s0 imp s0 2i 1 2i 2 푖 [ ( ) ( ) [ ( − )·( − )] − ] (− · ) ∑︁ ∑︁푘,푘0 푖 dp pd dp pd = Tr G k V G k q G k V G k q ∗ exp 푖q r , (S18) 2휋 [ s0 ( ) imp s0 ( − ) − [ s0 ( ) p s0 ( + )] ] (− · 1) ∑︁푘 and combining Eq. (S9), (S13), (S14) and (S15) we find 1 휌푠퐵 q = ImTr Gs0 2i 1, 1 VpGs0 1, 2i 1 exp 푖q r2i 1 (S19) 휋 + ( ) 푖 − [ ( + ) ( + )] (− · ) ∑︁ 푖 pp pp = Tr G k VpG k0 exp 푖 k k0 r2i 1 r1 푐.푐. exp 푖q r2i 1 (S20) 휋 s0 s0 + + 2 푖 [ ( ) ( ) [ ( − )·( − )] − ] (− · ) ∑︁ ∑︁푘,푘0 푖 pp pp pp pp = Tr G k V G k q G k V G k q ∗ exp 푖q r . (S21) 2휋 [ s0 ( ) p s0 ( − ) − [ s0 ( ) p s0 ( + )] ] (− · 1) ∑︁푘 To make the above expression more computationally tractable, we drop the common factor exp 푖q r for 휌푠퐴 and 휌푠퐵. (− · 1) Additionally, since Vimp = 푉I for an isotropic nonmagnetic impurity we find

푖 dp pd dp pd 휌푠퐴 q = 푉 Tr G k q G k Tr G k G k q ∗ (S22) ( ) 2휋 [ s0 ( + ) s0 ( )] − [ s0 ( ) 0 ( + )] ∑︁푘 푖 dp pd dp pd = 푉 Tr G k q G k Tr G k G k q ∗ (S23) 2휋 [ s0 ( + ) s0 ( )] − [ 0 ( ) s0 ( + )] ∑︁푘 and

푖 pp pp pp pp 휌푠퐵 q = 푉 Tr G k q G k Tr G k G k q ∗ (S24) ( ) 2휋 [ s0 ( + ) s0 ( )] − [ s0 ( ) s0 ( + )] ∑︁푘 1 pp pp = 푉 TrIm G k q G k . (S25) − 휋 [ s0 ( + ) s0 ( )] ∑︁푘 Finally, the total spectral function is

B q = 휌푠퐴 q 휌푠퐵 q . (S26) ( ) ( ) + ( ) In many cases, the off-diagonal contribution 휌푠퐴 is much smaller than 휌푠퐵. It is a good approximation to just keep only 휌푠퐵 q ( )) in 퐵 q [? ]. In our case, because 휌푠퐴 is comparable with 휌푠퐵 for some q, we still take the Eq. (S26) to calculate the QPI spectra. ( ) 3

SURFACE ELECTRONIC STRUCTURE AND QPI OF 001 -ORIENTED SURFACE ( )

In the following we compare our calculated surface spectral function and corresponding theoretically predicted QPI intensity with experimental results on the (001)-oriented surface, which is orthogonal to the (010)-oriented surface presented in the main manuscript. Due to technical limitations – the available in-plane magnetic field is limited to 3 Tesla – our QPI calculations and predictions of Fermi arcs on the (010)-oriented surface cannot be directly confirmed by experiments. However, the good agreement between our calculations and experiments on the (001)-oriented surface confirms and validates our methodology and calculations. We have calculated the surface spectral function and the Bi vacancy induced QPI spectra at 휔 = 100 meV and 200 meV. − − Figure S1 compares our theoretically obtained QPI spectra to experimental results. In our theoretical spectral functions [Fig. S1(a) and S1(d)] no Fermi arc is present due to the absence of 푥푦-mirror symmetry in the 001 -oriented slab. Rather, a trivial ( ) cross-like Fermi surface pocket surrounding 푀¯ and weak weight near Γ¯ is present. In the corresponding QPI maps [Fig. S1(b) and S1(e)] the main scattering peak around q = 0 is circular for 휔 = 100 meV, but transforms in to a rectangle for 휔 = 200 meV. − − In our experiments, using scanning tunneling spectroscopy (STS) we measured QPI on the 001 -oriented surface of CeBi in ( ) the FM-like state at 4 K (see Figs. S1 (c), (f)). Single crystals of CeBi were grown out of a Bi-rich binary melt with an initial composition of Ce26Bi74 [? ]. Elemental Ce and Bi were placed in a 3-cap-Ta crucible [? ] and slowly cooled from 1200 °C to 940 °C over 60 hours. Once at 940 °C the excess solution was decanted from the CeBi single crystals with the aid of a centrifuge [? ]. CeBi samples were then cleaved in cryogenic, ultrahigh vacuum at 30 K before inserting them into the STM head. The Ce magnetic moments are fully aligned by a 9 Tesla external magnetic fields applied along the 푐-axis. By overlaying the key features of our theoretical results on top of the experimental QPI spectra [Fig. S1(c) and S1(f)] we find reasonable agreement. For the case of 휔 = 100 meV, the theoretical and experimental results both have a circular high intensity peak around Γ¯. Theory − and experimental QPI is also consistent for the intense scattering near the corners of the Brillouin zone marked by a dashed red curve. For 휔 = 200 meV, we find the same level of agreement between theory and experiment, where the size of the predicted − and measured rectangle around Γ¯ almost exactly the same. By considering an isotropic nonmagnetic impurity potential at the Bi site, we obtain theoretical results in good accord with the experimentally observed QPI. This suggests that Bi vacancies at the surface of CeBi behave as isotropic nonmagnetic impurities. Therefore, the T-matrix approximation T = Vimp is reasonable and produces realistic results.

FIG. S1: Surface projected spectral function for (a) 휔 = 100 meV and (d) 휔 = 200 meV. QPI spectra for a single Bi impurity for (b) − − 휔 = 100 meV and (e) 휔 = 200 meV. Experimental QPI map measured at (c) 100mV and (f) 200mV bias voltages. Setup parameters for − − − − the QPI maps (c) and (f) are 푉푠 = 200 mV, 푅퐽 = 0.4 GΩ, where 푉푠 denotes the sample bias and 푅퐽 the tunneling junction resistance. The bias excitation amplitude was set to 푉rms = 7.1 mV. 4

TOPOLOGICAL NONTRIVIAL AND TRIVIAL SURFACE BANDS

To show the band dispersions related to the Fermi arcs, we calculated the band structures of a 30-layer (010) slab with respect to 푘 푥 for a list of 푘푧 as shown in Fig. S2. Both the open and closed boundary conditions are considered to distinguish the nontrivial surface band. We can see several features by comparing these panels. 1) Energy bands with linear dispersions are available in panel (a), (c), (e) and (g) but absent in (b), (d), (f) and (h). This means the existence of nontrivial massless bands at the open surface, which form the Fermi arcs. The bands have small gaps thus not perfectly massless because the finite depth of the slab. 2) We can see two linear-dispersion bands in (i) 푘푧 = 0.2 but only one in (k) 푘푧 = 0.25 , which indicates a Weyl node ( ) ( ) between 푘푧 = 0.2 and 0.25 because the 2D momentum plane between a pair of Weyl nodes has a nonzero Chern number [? ]. 3) For the same reason, one linear-dispersion band in (k) but none in (m) indicates another Weyl node between 푘푧 = 0.25 and 0.3. The energy cuts on bands in Fig. S2 correspond to the peaks in the map of spectral functions as shown in Fig. S3. Specifically, the cuts on the bands with linear dispersion form the Fermi acrs. Fig. S3(h)-(n) show the exchange of red and blue band (Fermi arcs) around the central area. This exchange is due to crossing of the massless linear bands shown in Fig. S2.

∗ Electronic address: [email protected] 5

FIG. S2: Band structures of a 30-layer (010) slab with open and closed boundary condition in the 푦-direction for various values of 푘푧. 푘푧 = 0, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35 and 0.4 for panels from the first to the last row. The boundary condition is open and closed for panels in the left and right column. 6

FIG. S3: (a) - (g) Surface spectral function of a (010)-oriented slab at 퐸 = 0.2, 0.15, 0.1, 0.05, 0, 0.05 and 0.1 eV. (h) - (n) Spectral − − − − functions for spin polarization 푆푧 at 퐸 = 0.2, 0.15, 0.1, 0.05, 0, 0.05 and 0.1 eV. h i − − − −